From owner-chemistry@ccl.net Sat Mar 16 17:38:01 2019 From: "Thomas Manz thomasamanz[a]gmail.com" To: CCL Subject: CCL: comment on double hybrids and DFT functional popularity Message-Id: <-53653-190316173355-24161-E8DG8hO/Hh3ZYIKaEDuuvA : server.ccl.net> X-Original-From: Thomas Manz Content-Type: multipart/alternative; boundary="0000000000003bde2e05843ce980" Date: Sat, 16 Mar 2019 15:33:37 -0600 MIME-Version: 1.0 Sent to CCL by: Thomas Manz [thomasamanz,+,gmail.com] --0000000000003bde2e05843ce980 Content-Type: text/plain; charset="UTF-8" Hi, In the extremely interesting article "A trip to the density functional zoo: Warnings and recommendations for the user" (DOI:10.1071/CH19023) that was recently brought to our attention by Susi ... Page I: "Dispersion-corrected, semi-empirical double hybrids are the most accurate DFT methods for ground-state thermochemistry." Pages G-H : "It is striking that even in the 2017 poll one third of the first division methods belong to the 16 worst DFA approaches for GMTKN55! We also note in passing that DHDFs, even though they have been shown to be the most robust DFT methods have not had a significant influence on the first division over the years." The article's authors are puzzled by why double hybrids (DHDFs) are not more popular despite their claimed high accuracy. I believe the reason double hybrids are not more popular is two-fold: (a) Double hybrids include a MP2-like perturbation energy as one of the components of their energy functional. The MP2 energy terms include a difference between occupied and virtual orbitals in their denominator. Metallic conductors (e.g., solids like Cu, Ag, Au, etc.) have a partially filled conduction band, so the energy difference between "occupied" and "virtual" orbitals is zero. This makes the MP2 energy blow up for these materials. Therefore, double hybrids are not suitable for computations on metallic conductors or other materials like graphene that have zero band-gap. Many large computational chemistry groups do research in which conducting materials play a large role. For example, heterogeneous catalysts containing metal particles on oxide supports. Many materials scientists and electrical engineers do research in which conducting materials interface with semi-conducting or insulating materials. Double hybrids are not applicable to these situations. (b) Double hybrids have a higher computational cost than GGA and hybrid functionals. If someone is studying a small molecule, then they can use higher level methods like coupled-cluster or configuration interaction calculations to get more accurate answers. If they are studying materials having a large number of atoms in the unit cell, then they are probably going to go for a less computationally expensive approach like a GGA or a hybrid functional. In my opinion, issue (a) is the more important one. The PBE functional is amazingly popular. PBE does not rank highest on many of the small molecule energy benchmarks, but it works well for conducting materials (e.g., pure transition metal solids) and it gives reasonable results for the magnetic properties of the 3-d pure transition metal solids (i.e., predicting the magnetic moments of chromium, iron, cobalt, & nickel quite well, and the non-magnetism of many other pure transition metal solids). It also does a reasonable job of predicting the magnetic anisotropy barrier of the Mn12-acetate single molecule magnet, though its results (like many other functionals) are somewhat hit or miss for predicting the ground state magnetic properties of many organometallic compounds. When a dispersion correction is included (e.g., PBE+dispersion), then PBE often does an excellent job for the cohesive energies, bulk moduli, and lattice constants of diverse solids. PBE is obviously not perfect. It fails spectacularly for many highly correlated materials and often gives inaccurate band gaps for semi-conductors and insulators. Going forward a lot of the need for new DFT functionals concerns the highly correlated materials such as complex oxides and complicated magnetic materials. To the best of my knowledge, all existing DFT functionals do not provide a thoroughly satisfying description of highly correlated materials. Why is there so much focus within the community of developing new DFT functionals on small molecule thermochemistry? That is only a small part of the field of computational chemistry. Huge impact could be made if someone solves the larger problem of giving consistently reasonable results for complex materials. In my opinion, articles proposing new DFT functionals should get more creative in their test sets. In addition to reporting accuracy for thermochemistry, they should quantify convergence reliability, computational cost, and performance across more diverse material types. They should place more focus predicting correct ground spin states. For example, to include test sets of the ground spin/magnetic state prediction for isolated atoms, organometallic complexes, complex oxide solids, lanthanide and actinide compounds, endohedral complexes, etc. from available experimental data. There are many fine experimental papers where people have used advanced techniques like EPR spectroscopy, Mossbauer spectroscopy, polarized neutron diffraction, and other spectroscopic techniques to get the ground magnetic state data. This test data should include a cross-section of conducting, semi-conducting, insulating, and half-metal systems. In addition to the magnetic alignment, quantification of atomic spin moments and magnetic anisotropy barriers could be included for some materials where these have been experimentally measured. Then we could start to address the real issues of how a DFT functional performs for complex materials. Sincerely, Tom --0000000000003bde2e05843ce980 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable
Hi,

In the extremely interesting articl= e "A trip to the density functional zoo: Warnings and recommendations = for the user" (DOI:10.1071/CH19023) that was recently brought to our a= ttention by Susi ...

Page I: "Dispersion-corr= ected, semi-empirical double hybrids are the most accurate DFT methods for = ground-state thermochemistry."
Pages G-H : "It is strik= ing that even in the 2017 poll one third of the first division methods belo= ng to the 16 worst DFA approaches for GMTKN55! We also note in passing that= DHDFs, even though they have been shown to be the most robust DFT methods = have not had a significant influence on the first division over the years.&= quot;

The article's authors are puzzled by why= double hybrids (DHDFs) are not more popular despite their claimed high acc= uracy.

I believe the reason double hybrids are not= more popular is two-fold:

(a) Double hybrids incl= ude a MP2-like perturbation energy as one of the components of their energy= functional. The MP2 energy terms include a difference between occupied and= virtual orbitals in their denominator. Metallic conductors (e.g., solids l= ike Cu, Ag, Au, etc.) have a partially filled conduction band, so the energ= y difference between "occupied" and "virtual" orbitals = is zero. This makes the MP2 energy blow up for these materials. Therefore, = double hybrids are not suitable for computations on metallic conductors or = other materials like graphene that have zero band-gap.

=
Many large computational chemistry groups do research in which conduct= ing materials play a large role. For example, heterogeneous catalysts conta= ining metal particles on oxide supports. Many materials scientists and elec= trical engineers do research in which conducting materials interface with s= emi-conducting or insulating materials. Double hybrids are not applicable t= o these situations.

(b) Double hybrids have a high= er computational cost than GGA and hybrid functionals. If someone is studyi= ng a small molecule, then they can use higher level methods like coupled-cl= uster or configuration interaction calculations to get more accurate answer= s. If they are studying materials having a large number of atoms in the uni= t cell, then they are probably going to go for a less computationally expen= sive approach like a GGA or a hybrid functional.=C2=A0

=
In my opinion, issue (a) is the more important one.


The PBE functional is amazingly popular. PBE does not= rank highest on many of the small molecule energy benchmarks, but it works= well for conducting materials (e.g., pure transition metal solids) and it = gives reasonable results for the magnetic properties of the 3-d pure transi= tion metal solids (i.e., predicting the magnetic moments of chromium, iron,= cobalt, & nickel quite well, and the non-magnetism of many other pure = transition metal solids). It also does a reasonable job of predicting the m= agnetic anisotropy barrier of the Mn12-acetate single molecule magnet, thou= gh its results (like many other functionals) are somewhat hit or miss for p= redicting the ground state magnetic properties of many organometallic compo= unds. When a dispersion correction is included (e.g., PBE+dispersion), then= PBE often does an excellent job for the cohesive energies, bulk moduli, an= d lattice constants of diverse solids. PBE is obviously not perfect. It fai= ls spectacularly for many highly correlated materials and often gives inacc= urate band gaps for semi-conductors and insulators.

Going forward a lot of the need for new DFT functionals concerns the high= ly correlated materials such as complex oxides and complicated magnetic mat= erials. To the best of my knowledge, all existing DFT functionals do not pr= ovide a thoroughly satisfying description of highly correlated materials. W= hy is there so much focus within the community of developing new DFT functi= onals on small molecule thermochemistry? That is only a small part of the f= ield of computational chemistry. Huge impact could be made if someone solve= s the larger problem of giving consistently reasonable results for complex = materials.=C2=A0

In my opinion, articles proposing= new DFT functionals should get more creative in their test sets. In additi= on to reporting accuracy for thermochemistry, they should quantify converge= nce reliability, computational cost, and performance across more diverse ma= terial types. They should place more focus predicting correct ground spin s= tates. For example, to include test sets of the ground spin/magnetic state = prediction for isolated atoms, organometallic complexes, complex oxide soli= ds, lanthanide and actinide compounds, endohedral complexes, etc. from avai= lable experimental data. There are many fine experimental papers where peop= le have used advanced techniques like EPR spectroscopy, Mossbauer spectrosc= opy, polarized neutron diffraction, and other spectroscopic techniques to g= et the ground magnetic state data. This test data should include a cross-se= ction of conducting, semi-conducting, insulating, and half-metal systems. I= n addition to the magnetic alignment, quantification of atomic spin moments= and magnetic anisotropy barriers could be included for some materials wher= e these have been experimentally measured.=C2=A0Then we could start to addr= ess the real issues of how a DFT functional performs for complex materials.=

Sincerely,

Tom=C2=A0
--0000000000003bde2e05843ce980-- From owner-chemistry@ccl.net Sat Mar 16 19:27:00 2019 From: "Kenneth Ruud kenneth.ruud{:}uit.no" To: CCL Subject: CCL: Second announcement: ISTCP-X, Troms July 11-17 2019 Message-Id: <-53654-190316185853-21472-SESJElj9hfEjHMejwvoi2A[]server.ccl.net> X-Original-From: "Kenneth Ruud" Date: Sat, 16 Mar 2019 18:58:49 -0400 Sent to CCL by: "Kenneth Ruud" [kenneth.ruud/a\uit.no] We are happy to announce that a first complete draft of the program for the 10th Congress of the International Society of Theoretical Chemical Physical is now available on our web pages (http://istcp-2019.org/program.html). Please note that the dates and times for individual speakers may still change. - Early-bird registration deadline ends on April 30 2019. - Deadline for booking rooms at the official conference hotel is April 30 2019. - Deadline for submission of poster abstracts is June 16 2019. - Final deadline for registration for the conference is June 16 2019. Please visit the conference web pages at http://istcp-2019.org for more information, registration and abstract submission. The ISTCP Congresses showcase the achievements and advances of all areas of theoretical chemical physics, with special emphasis on the interaction between experimental and theoretical physical chemistry, chemical physics, materials and life sciences. This years congress will feature 202 invited speakers. There are 12 plenary speakers, whereas the rest of the presentations will be given in parallel sessions organized around 14 scientific tracks: - From picoseconds to attoseconds: nuclear and electron dynamics (Chairs: David Clary, Leticia Gonzalez and Fernando Martin) - Aspects of heavy-element chemistry (Chairs: Pekka Pyykko and Trond Saue) - Emergent electronic structure methods (Chairs: Stefan Goedecker and Gustavo Scuseria) - Multiscale modeling including focused models (Chairs: Benedetta Mennucci and Lyudmila Slipchenko) - Large-scale electronic structure models of materials (Chairs: Thomas Heine and Hiromi Nakai) - Ultracold chemical physics (Chairs: Jeremy M. Hutson and Bogumil Jeziorski) - Molecular Properties and interactions (Chairs: Antonio Rizzo and Krzysztof Szalewicz) - Computational spectroscopy: from x-rays to microwaves (Chairs: Attila Csaszar and Hans Agren) - 90 years of r12: Hylleraas symposium (Chairs: Wim Klopper and Ed Valeev) - Machine learning and data-driven approaches in chemical physics (Chairs: Alan Aspuru-Guzik and Pavlo Dral) - Computational biophysics (Chairs: Fernanda Duarte Gonzalez and Ursula Rothlisberger) - Path-integral methods (Chairs: David E. Manolopoulos and Gregory Voth) - Physical organic chemistry and catalysis (Chairs: Odile Eisenstein and Vidar R. Jensen) - Janos Ladik memorial symposium (Chairs: Erkki Brandas and Kenneth Ruud) The 12 plenary/keynote speakers are: Irene Burghardt (University of Frankfurt, Germany) Sylvia Canuto (University of Sao Paulo, Brazil) Monica Olvera de la Cruz (Northwestern University, Evanston (IL), USA) Guilia Galli (University of Chicago (IL), USA) Sharon Hammes-Schiffer (Yale, New Haven (CT), USA) Trygve Helgaker (University of Oslo, Norway) Kersti Hermansson (University of Uppsala, Sweden) Thomas F. Miller (CalTech, Los Angeles (CA), USA) Peter Saalfrank (University of Potsdam, Germany) Peter Schwerdtfeger (Massey University, Auckland, New Zealand) Birgitta Whaley (University of California Berkeley (CA), USA) Zhigang Shuai (Tsinghua University, Beijing, China) For the complete list of speakers, please see our web pages: http://istcp- 2019.org/speakers_session.html On behalf of the organizing committees, Kenneth Prof. Kenneth Ruud Chair, ISTCP-X UiT The Arctic University of Norway https://uit.no/go/target/41020 E-mail: kenneth.ruud- -uit.no Telephone/mobile: +47 77623101/+47 90098353